Add an agent session driver and Gadget durability tests - #270
Draft
AshishKumar4 wants to merge 7 commits into
Draft
Add an agent session driver and Gadget durability tests#270AshishKumar4 wants to merge 7 commits into
AshishKumar4 wants to merge 7 commits into
Conversation
This was referenced Aug 19, 2026
Maximo-Guk
reviewed
Aug 19, 2026
AgentSession drives one agent session over the same Cap'n Web API the browser uses: a fresh account and workspace, one chat across turns, complete paginated history, and an optional source snapshot. Two methods support tests that need a known implementation rather than whatever an agent produced. seedGadget() writes hand-authored source into the workspace. restartGadgets() restarts every Gadget server by applying an empty code update, which is what the platform does on every code change. gadget-durability.test.ts uses both to pin platform behaviour with no model involved. Storage survives a restart and memory does not, outstanding stubs become invalid, and the data holds across five restarts and across one that interrupts a write. It also shows that a check-then-write implementation oversells under concurrent calls. startHarness() gains enableGadgetExecution, which keeps the Worker Loader so Gadget code can run. It defaults to false, so the existing suites are unchanged.
A handler receives the URL, the method, and the headers, but never the body, so it cannot stand in for a host a suite has to reach with a real POST. passThroughHosts exempts such a host before the request is taken apart. Every other host still throws.
Every .js file in a Gadget becomes a module in its Worker, so workerd parses client.js at load even though the server never imports it. A test that checks a Gadget through its RPC therefore already covers the syntax of both files, and a separate parse step would add nothing.
AshishKumar4
force-pushed
the
evals/toolkit
branch
from
August 21, 2026 16:56
641a57e to
bf05017
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds a driver for one agent session, and integration tests that pin how a Gadget server behaves when
it restarts.
Everything here is deterministic and needs no model. It runs in
pnpm testand finishes in about tenseconds.
AgentSession
AgentSessiondrives one session over the same Cap'n Web API that the browser uses. It creates a freshaccount and workspace, keeps one chat across turns, reads the complete paginated history, and can
snapshot the accepted source.
Two methods support a test that needs a known implementation:
seedGadget({title, bindingName, files})writes hand-authored source into the workspace. It createsa Gadget without the agent.
restartGadgets()restarts every Gadget server. It applies an empty update to the mainline code,which is what the platform does on every code change.
startHarness()gainsenableGadgetExecution, which keeps the Worker Loader so Gadget code can run.It defaults to false, so the existing suites do not change.
What the tests pin
__tests__/gadget-durability.test.tsuses a Gadget that holds one value in SQL and one in memory:The last item pins the behaviour that any "never oversells" assertion depends on.
Scope
This is the first of two changes. The evaluation suite that uses this driver is in
#271.